[LU-4627] Client deadlock on ll_setattr_raw

• Type: Bug
• Resolution: Fixed
• Priority: Blocker
• Fix Version/s: Lustre 2.6.0, Lustre 2.5.4
• Affects Version/s: Lustre 2.5.0
• Severity: 3
• Assignee: Zhenyu Xu
• Reporter: Henri Doreau (Inactive)

Description

While investigating LU-3732 I've regularly bumped into what seems to be a deadlock on lli_trunc_sem, according to the trace below:

Call Trace:
[<ffffffff81511655>] rwsem_down_failed_common+0x95/0x1d0
[<ffffffff815117b3>] rwsem_down_write_failed+0x23/0x30
[<ffffffff81284163>] call_rwsem_down_write_failed+0x13/0x20
[<ffffffff81510cb2>] ? down_write+0x32/0x40
[<ffffffffa0e03fe1>] ll_setattr_raw+0x191/0x10c0 [lustre]
[<ffffffff810758c7>] ? current_fs_time+0x27/0x30
[<ffffffffa0e04f6d>] ll_setattr+0x5d/0xf0 [lustre]
[<ffffffff8119ec18>] notify_change+0x168/0x340
[<ffffffffa06bb3bb>] ? libcfs_debug_vmsg2+0x50b/0xbb0 [libcfs]
[<ffffffff811197ef>] file_remove_suid+0x5f/0x90
[<ffffffff8111c2c0>] __generic_file_aio_write+0x220/0x490
[<ffffffffa06bbaa1>] ? libcfs_debug_msg+0x41/0x50 [libcfs]
[<ffffffff8111c5b8>] generic_file_aio_write+0x88/0x100
[<ffffffffa0e4353b>] vvp_io_write_start+0xdb/0x3d0 [lustre]
[<ffffffffa08cbcaa>] cl_io_start+0x6a/0x140 [obdclass]
[<ffffffffa08cfe24>] cl_io_loop+0xb4/0x1b0 [obdclass]
[<ffffffffa0de2a36>] ll_file_io_generic+0x2b6/0x710 [lustre]
[<ffffffffa08bfdf9>] ? cl_env_get+0x29/0x350 [obdclass]
[<ffffffffa0de3702>] ll_file_aio_write+0x142/0x2c0 [lustre]
[<ffffffffa0de39ec>] ll_file_write+0x16c/0x2a0 [lustre]
[<ffffffff811814b8>] vfs_write+0xb8/0x1a0
[<ffffffff81181e72>] sys_pwrite64+0x82/0xa0
[<ffffffff8100b072>] system_call_fastpath+0x16/0x1b

It seems to me that the down_read() in ll_file_io_generic() should only be done if iot == CIT_READ since in the write case, we down_write() later.

Issue Links

• is duplicated by: LU-4710 Deadlock on lli_trunc_sem in ll_setattr_raw()